A complete speaking model
In education, AI can suggest practice and provide quick feedback, but a teacher should check important decisions and students should be able to challenge errors. I would use a system to identify a possible difficulty, then ask the learner to explain. Imagine a vocabulary answer that disappears during a connection failure. A blank saved record does not prove that the learner does not know the word. The system should request clarification or another attempt, not attach a permanent low-ability label. It should also avoid collecting private information that the practice task does not require. Teachers and learners need a clear way to report doubtful feedback and see why a recommendation was made. A score based on missing data is not a fair judgement. Used this way, an automated tool supports learning while responsibility stays with people.
Follow-up: Should a confident machine answer end the discussion?
Response: No. Check the evidence and allow a human review, especially when the result affects a learner.